Activities Of Daily Living Recognition And Abnormal Detection For The Elderly Based On Smart Home Sensor Data

With the aggravation of population ageing and the increase of human resources cost,endowment has become the main topic of current society faces.Automatically recognizing the acticities of daily living of the elderly and discovering the abnormal is a feasible way to slove old-age life problems,so as to improve the elderly's independent living ability and level of family health care,For the sake of recognizing the elderly's activities,understanding the elderly's intentions and discovering the abnormal of activities and environment,this paper combines the theory of machine learning and data processing technology with smart home technology based on the experimental data of Washington State University's Laboratory of Intelligent Space.Main research contents as follows:Firstly,a method for dynamic segmentation of sensor event is proposed.Based on the sensors dependency,similarity degree of neighboring windows is calculated.Then finding and saving the points which have the lowest similarity in continuous borders.Finaly continuous sequence segments with identical behavior label are merged.Secondly,a model which can recognise various daily acticities with high accuarcy is constructed.According to the daily behavior characteristics of the elderly,multidimensional feature frame includes sensor weight,sensor class,time and frequency.is constructed.In order to establish the daily activity model,genetic algorithm and crossvalidation are used to train and adjust the model,Thirdly,designing a frame to detecting abnormal behavior of the elderly.Using data of the most important features to train gaussian mixture model for every activity.Using the gaussian model and SVM model to detecte abnormal through calculate the deviation of abnormal samples and their components.Then analysing the differences of features between normal and abnormal acitvities.Recognising the elderly's daily activities is premise conditions of providing living assistance for the aged,improving their independent living ability and level of family health care.The effectiveness of the method and model proposed in this paper was validated.The experimental results indicate that the sequence segmentation method can segment sensor sequences reasonably,and the daily activities model can identify various acitivity accurately.The abnormal detection method can find out the abnormality of the elderly's behavior effectively.The results of this paper will be helpful to promote the development of aged home-based community caring.